FIO Windows update
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index debcac5eea9edeb55eea6ff0488e7389f8a5de14..93482f5294ed55c245e25a0e98f8151fe7354ced 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -28,6 +28,7 @@
 #include <time.h>
 #include <locale.h>
 #include <assert.h>
+#include <time.h>
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <sys/ipc.h>
 #include "hash.h"
 #include "smalloc.h"
 #include "verify.h"
+#include "trim.h"
 #include "diskutil.h"
+#include "cgroup.h"
+#include "profile.h"
+#include "lib/rand.h"
 
 unsigned long page_mask;
 unsigned long page_size;
@@ -58,8 +63,10 @@ static struct fio_mutex *startup_mutex;
 static struct fio_mutex *writeout_mutex;
 static volatile int fio_abort;
 static int exit_value;
-static struct itimerval itimer;
+static timer_t ival_timer;
 static pthread_t gtod_thread;
+static struct flist_head *cgroup_list;
+static char *cgroup_mnt;
 
 struct io_log *agg_io_log[2];
 
@@ -94,12 +101,12 @@ static void terminate_threads(int group_id)
                         * if the thread is running, just let it exit
                         */
                        if (td->runstate < TD_RUNNING)
-                               kill(td->pid, SIGQUIT);
+                               kill(td->pid, SIGTERM);
                        else {
                                struct ioengine_ops *ops = td->io_ops;
 
-                               if (ops && (ops->flags & FIO_SIGQUIT))
-                                       kill(td->pid, SIGQUIT);
+                               if (ops && (ops->flags & FIO_SIGTERM))
+                                       kill(td->pid, SIGTERM);
                        }
                }
        }
@@ -107,17 +114,21 @@ static void terminate_threads(int group_id)
 
 static void status_timer_arm(void)
 {
-       itimer.it_value.tv_sec = 0;
-       itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
-       setitimer(ITIMER_REAL, &itimer, NULL);
+       struct itimerspec value;
+
+       value.it_value.tv_sec = 0;
+       value.it_value.tv_nsec = DISK_UTIL_MSEC * 1000000;
+       value.it_interval.tv_sec = 0;
+       value.it_interval.tv_nsec = DISK_UTIL_MSEC * 1000000;
+
+       timer_settime(ival_timer, 0, &value, NULL);
 }
 
-static void sig_alrm(int fio_unused sig)
+static void ival_fn(union sigval sig)
 {
        if (threads) {
                update_io_ticks();
                print_thread_status();
-               status_timer_arm();
        }
 }
 
@@ -131,46 +142,43 @@ static void sig_quit(int sig)
 static void sig_int(int sig)
 {
        if (threads) {
-               printf("\nfio: terminating on signal %d\n", sig);
+               log_info("\nfio: terminating on signal %d\n", sig);
                fflush(stdout);
                terminate_threads(TERMINATE_ALL);
        }
 }
 
-static void sig_ill(int fio_unused sig)
+static void posix_timer_teardown(void)
 {
-       if (!threads)
-               return;
+       timer_delete(ival_timer);
+}
+
+static void posix_timer_setup(void)
+{
+       struct sigevent evt;
+
+       memset(&evt, 0, sizeof(evt));
+       evt.sigev_notify = SIGEV_THREAD;
+       evt.sigev_notify_function = ival_fn;
+
+       if (timer_create(FIO_TIMER_CLOCK, &evt, &ival_timer) < 0)
+               perror("timer_create");
 
-       log_err("fio: illegal instruction. your cpu does not support "
-               "the sse4.2 instruction for crc32c\n");
-       terminate_threads(TERMINATE_ALL);
-       exit(4);
 }
 
 static void set_sig_handlers(void)
 {
        struct sigaction act;
 
-       memset(&act, 0, sizeof(act));
-       act.sa_handler = sig_alrm;
-       act.sa_flags = SA_RESTART;
-       sigaction(SIGALRM, &act, NULL);
-
        memset(&act, 0, sizeof(act));
        act.sa_handler = sig_int;
        act.sa_flags = SA_RESTART;
        sigaction(SIGINT, &act, NULL);
 
-       memset(&act, 0, sizeof(act));
-       act.sa_handler = sig_ill;
-       act.sa_flags = SA_RESTART;
-       sigaction(SIGILL, &act, NULL);
-
        memset(&act, 0, sizeof(act));
        act.sa_handler = sig_quit;
        act.sa_flags = SA_RESTART;
-       sigaction(SIGQUIT, &act, NULL);
+       sigaction(SIGTERM, &act, NULL);
 }
 
 /*
@@ -187,6 +195,8 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
        unsigned int rate_iops = 0;
        unsigned int rate_iops_min = 0;
 
+       assert(ddir_rw(ddir));
+
        if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
                return 0;
 
@@ -429,6 +439,8 @@ static void do_verify(struct thread_data *td)
        int ret, min_events;
        unsigned int i;
 
+       dprint(FD_VERIFY, "starting loop\n");
+
        /*
         * sync io first and invalidate cache, to make sure we really
         * read from disk.
@@ -500,7 +512,8 @@ static void do_verify(struct thread_data *td)
                                io_u->xfer_buf += bytes;
                                io_u->offset += bytes;
 
-                               td->ts.short_io_u[io_u->ddir]++;
+                               if (ddir_rw(io_u->ddir))
+                                       td->ts.short_io_u[io_u->ddir]++;
 
                                if (io_u->offset == f->real_file_size)
                                        goto sync_done;
@@ -536,7 +549,8 @@ sync_done:
                 */
                full = queue_full(td) || ret == FIO_Q_BUSY;
                if (full || !td->o.iodepth_batch_complete) {
-                       min_events = td->o.iodepth_batch_complete;
+                       min_events = min(td->o.iodepth_batch_complete,
+                                        td->cur_depth);
                        if (full && !min_events)
                                min_events = 1;
 
@@ -565,6 +579,8 @@ sync_done:
                cleanup_pending_aio(td);
 
        td_set_runstate(td, TD_RUNNING);
+
+       dprint(FD_VERIFY, "exiting loop\n");
 }
 
 /*
@@ -582,6 +598,7 @@ static void do_io(struct thread_data *td)
                td_set_runstate(td, TD_RUNNING);
 
        while ( (td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
+               (!flist_empty(&td->trim_list)) ||
                ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) ) {
                struct timeval comp_time;
                unsigned long bytes_done[2] = { 0, 0 };
@@ -642,7 +659,8 @@ static void do_io(struct thread_data *td)
                                io_u->xfer_buf += bytes;
                                io_u->offset += bytes;
 
-                               td->ts.short_io_u[io_u->ddir]++;
+                               if (ddir_rw(io_u->ddir))
+                                       td->ts.short_io_u[io_u->ddir]++;
 
                                if (io_u->offset == f->real_file_size)
                                        goto sync_done;
@@ -688,7 +706,8 @@ sync_done:
                 */
                full = queue_full(td) || ret == FIO_Q_BUSY;
                if (full || !td->o.iodepth_batch_complete) {
-                       min_evts = td->o.iodepth_batch_complete;
+                       min_evts = min(td->o.iodepth_batch_complete,
+                                       td->cur_depth);
                        if (full && !min_evts)
                                min_evts = 1;
 
@@ -735,6 +754,9 @@ sync_done:
                }
        }
 
+       if (td->trim_entries)
+               printf("trim entries %ld\n", td->trim_entries);
+
        if (td->o.fill_device && td->error == ENOSPC) {
                td->error = 0;
                td->terminate = 1;
@@ -807,7 +829,8 @@ static int init_io_u(struct thread_data *td)
        if (allocate_io_mem(td))
                return 1;
 
-       if (td->o.odirect || td->o.mem_align)
+       if (td->o.odirect || td->o.mem_align ||
+           (td->io_ops->flags & FIO_RAWIO))
                p = PAGE_ALIGN(td->orig_buffer) + td->o.mem_align;
        else
                p = td->orig_buffer;
@@ -836,6 +859,13 @@ static int init_io_u(struct thread_data *td)
 
                        if (td_write(td) && !td->o.refill_buffers)
                                io_u_fill_buffer(td, io_u, max_bs);
+                       else if (td_write(td) && td->o.verify_pattern_bytes) {
+                               /*
+                                * Fill the buffer with the pattern if we are
+                                * going to be doing writes.
+                                */
+                               fill_pattern(td, io_u->buf, max_bs, io_u);
+                       }
                }
 
                io_u->index = i;
@@ -960,6 +990,8 @@ void reset_all_stats(struct thread_data *td)
        }
        
        fio_gettime(&tv, NULL);
+       td->ts.runtime[0] = 0;
+       td->ts.runtime[1] = 0;
        memcpy(&td->epoch, &tv, sizeof(tv));
        memcpy(&td->start, &tv, sizeof(tv));
 }
@@ -998,7 +1030,7 @@ static int exec_string(const char *string)
  */
 static void *thread_main(void *data)
 {
-       unsigned long long runtime[2], elapsed;
+       unsigned long long elapsed;
        struct thread_data *td = data;
        pthread_condattr_t attr;
        int clear_state;
@@ -1016,6 +1048,7 @@ static void *thread_main(void *data)
        INIT_FLIST_HEAD(&td->io_log_list);
        INIT_FLIST_HEAD(&td->io_hist_list);
        INIT_FLIST_HEAD(&td->verify_list);
+       INIT_FLIST_HEAD(&td->trim_list);
        pthread_mutex_init(&td->io_u_lock, NULL);
        td->io_hist_tree = RB_ROOT;
 
@@ -1036,6 +1069,15 @@ static void *thread_main(void *data)
         */
        fio_mutex_remove(td->mutex);
 
+       if (td->o.uid != -1U && setuid(td->o.uid)) {
+               td_verror(td, errno, "setuid");
+               goto err;
+       }
+       if (td->o.gid != -1U && setgid(td->o.gid)) {
+               td_verror(td, errno, "setgid");
+               goto err;
+       }
+
        /*
         * May alter parameters that init_io_u() will use, so we need to
         * do this first.
@@ -1073,6 +1115,9 @@ static void *thread_main(void *data)
                }
        }
 
+       if (td->o.cgroup_weight && cgroup_setup(td, cgroup_list, &cgroup_mnt))
+               goto err;
+
        if (nice(td->o.nice) == -1) {
                td_verror(td, errno, "nice");
                goto err;
@@ -1103,7 +1148,6 @@ static void *thread_main(void *data)
        fio_gettime(&td->epoch, NULL);
        getrusage(RUSAGE_SELF, &td->ts.ru_start);
 
-       runtime[0] = runtime[1] = 0;
        clear_state = 0;
        while (keep_running(td)) {
                fio_gettime(&td->start, NULL);
@@ -1128,11 +1172,11 @@ static void *thread_main(void *data)
 
                if (td_read(td) && td->io_bytes[DDIR_READ]) {
                        elapsed = utime_since_now(&td->start);
-                       runtime[DDIR_READ] += elapsed;
+                       td->ts.runtime[DDIR_READ] += elapsed;
                }
                if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
                        elapsed = utime_since_now(&td->start);
-                       runtime[DDIR_WRITE] += elapsed;
+                       td->ts.runtime[DDIR_WRITE] += elapsed;
                }
 
                if (td->error || td->terminate)
@@ -1149,15 +1193,15 @@ static void *thread_main(void *data)
 
                do_verify(td);
 
-               runtime[DDIR_READ] += utime_since_now(&td->start);
+               td->ts.runtime[DDIR_READ] += utime_since_now(&td->start);
 
                if (td->error || td->terminate)
                        break;
        }
 
        update_rusage_stat(td);
-       td->ts.runtime[0] = (runtime[0] + 999) / 1000;
-       td->ts.runtime[1] = (runtime[1] + 999) / 1000;
+       td->ts.runtime[0] = (td->ts.runtime[0] + 999) / 1000;
+       td->ts.runtime[1] = (td->ts.runtime[1] + 999) / 1000;
        td->ts.total_run_time = mtime_since_now(&td->epoch);
        td->ts.io_bytes[0] = td->io_bytes[0];
        td->ts.io_bytes[1] = td->io_bytes[1];
@@ -1170,6 +1214,13 @@ static void *thread_main(void *data)
                } else
                        finish_log(td, td->ts.bw_log, "bw");
        }
+       if (td->ts.lat_log) {
+               if (td->o.lat_log_file) {
+                       finish_log_named(td, td->ts.lat_log,
+                                               td->o.lat_log_file, "lat");
+               } else
+                       finish_log(td, td->ts.lat_log, "lat");
+       }
        if (td->ts.slat_log) {
                if (td->o.lat_log_file) {
                        finish_log_named(td, td->ts.slat_log,
@@ -1193,7 +1244,7 @@ static void *thread_main(void *data)
 
 err:
        if (td->error)
-               printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
+               log_info("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
                                                        td->verror);
 
        if (td->o.verify_async)
@@ -1202,6 +1253,7 @@ err:
        close_and_free_files(td);
        close_ioengine(td);
        cleanup_io_u(td);
+       cgroup_shutdown(td, &cgroup_mnt);
 
        if (td->o.cpumask_set) {
                int ret = fio_cpuset_exit(&td->o.cpumask);
@@ -1301,7 +1353,7 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                        if (WIFSIGNALED(status)) {
                                int sig = WTERMSIG(status);
 
-                               if (sig != SIGQUIT)
+                               if (sig != SIGTERM)
                                        log_err("fio: pid=%d, got signal=%d\n",
                                                        (int) td->pid, sig);
                                td_set_runstate(td, TD_REAPED);
@@ -1358,9 +1410,13 @@ static void *gtod_thread_main(void *data)
 
 static int fio_start_gtod_thread(void)
 {
+       pthread_attr_t attr;
        int ret;
 
-       ret = pthread_create(&gtod_thread, NULL, gtod_thread_main, NULL);
+       pthread_attr_init(&attr);
+       pthread_attr_setstacksize(&attr, PTHREAD_STACK_MIN);
+       ret = pthread_create(&gtod_thread, &attr, gtod_thread_main, NULL);
+       pthread_attr_destroy(&attr);
        if (ret) {
                log_err("Can't create gtod thread: %s\n", strerror(ret));
                return 1;
@@ -1394,17 +1450,17 @@ static void run_threads(void)
                return;
 
        if (!terse_output) {
-               printf("Starting ");
+               log_info("Starting ");
                if (nr_thread)
-                       printf("%d thread%s", nr_thread,
+                       log_info("%d thread%s", nr_thread,
                                                nr_thread > 1 ? "s" : "");
                if (nr_process) {
                        if (nr_thread)
                                printf(" and ");
-                       printf("%d process%s", nr_process,
+                       log_info("%d process%s", nr_process,
                                                nr_process > 1 ? "es" : "");
                }
-               printf("\n");
+               log_info("\n");
                fflush(stdout);
        }
 
@@ -1613,6 +1669,7 @@ int main(int argc, char *argv[])
        long ps;
 
        sinit();
+       init_rand(&__fio_rand_state);
 
        /*
         * We need locale for number printing, if it isn't set then just
@@ -1635,6 +1692,9 @@ int main(int argc, char *argv[])
        if (parse_options(argc, argv))
                return 1;
 
+       if (exec_profile && load_profile(exec_profile))
+               return 1;
+
        if (!thread_number)
                return 0;
 
@@ -1644,12 +1704,20 @@ int main(int argc, char *argv[])
        }
 
        startup_mutex = fio_mutex_init(0);
+       if (startup_mutex == NULL)
+               return 1;
        writeout_mutex = fio_mutex_init(1);
+       if (writeout_mutex == NULL)
+               return 1;
 
        set_genesis_time();
 
+       posix_timer_setup();
        status_timer_arm();
 
+       cgroup_list = smalloc(sizeof(*cgroup_list));
+       INIT_FLIST_HEAD(cgroup_list);
+
        run_threads();
 
        if (!fio_abort) {
@@ -1661,6 +1729,11 @@ int main(int argc, char *argv[])
                }
        }
 
+       cgroup_kill(cgroup_list);
+       sfree(cgroup_list);
+       sfree(cgroup_mnt);
+
+       posix_timer_teardown();
        fio_mutex_remove(startup_mutex);
        fio_mutex_remove(writeout_mutex);
        return exit_value;